In partnership with
November 10, 2026 · READ ONLINE
Welcome back, AI enthusiasts.
Every few weeks a new AI model takes the top spot, and this time Anthropic’s model built its own operating system and helped crack two black hole problems. Fable 5.2 is live for anyone with Claude, company gets ready for IPO on Wall Street. Let’s Dive In!
IN TODAY'S ISSUE
01 Anthropic’s Just Dropped Fable and Mythos 5.2
02 Google fires back at Nvidia with 2 NEW chips
03 Meta matches Claude at ONE-THIRD the price
01 / MODEL LAUNCH
Evolving AI: Anthropic dropped Fable 5.2 days before its IPO, and it beats GPT-6 Astra & Gemini 4 Carbon on most tests.
KEY POINTS
Fable 5.2 ran 1,000+ AI agents for 8 days and came back with an operating system that boots on a laptop.
Claude Fable 5.2 solved two open problems about black holes in theoretical physics.
Mythos 5.2 predicted the shape of 28 out of 34 gene therapy delivery shells on hardware a single scientist can run.
DETAILS
Anthropic launched Fable 5.2 for every Claude user, and it answers up to 30% faster than Fable 5.1. Mythos 5.2 is the same model without the hacking and biology filters, so only vetted teams can use it. METR, one of the model’s outside testers, found Anthropic’s AI progress has roughly doubled, likely because its own AI now speeds up the work. Fable 5.2 also costs less at $8 per million input tokens, and paid plans get extra Fable usage as a launch bonus.
WHY IT MATTERS
Wall Street gets a vote on how fast Anthropic moves once its shares start trading, and shareholders punish any quarter where growth slows down. That pressure stacks on top of a lab that is already speeding itself up, so the next wave of change in your job will arrive on an earnings schedule whether anyone is ready or not.
AI is only as good as what you tell it
The fastest way to give AI more context is to talk. But that only works if the transcription is right. Wispr Flow runs on Canto, its own speech model, trained on real dictation instead of studio audio.
In Wispr's tests on 10 hours of real dictations from 2,300+ speakers, Canto had the lowest word error rate of every model tested, including Google and OpenAI. It also pulls from your personal dictionary, so tool names, project names and teammates come out spelled right.
Talk through the whole problem, hit enter, and get a better answer. 4x faster than typing, in every app.
02 / CHIP ADVANCEMENT
Evolving AI: Google dropped two new AI chips aimed straight at the market Nvidia rules.
KEY POINTS
TPU 8i Max fuses two chips into one package with double the computing power and memory speed.
TPU 8tx keeps data and AI models protected even while they run, aimed at banks and governments.
PyTorch and Hugging Face models now run on Google’s chips with no code rewrites.
DETAILS
Google added the chips to its eighth TPU generation, which already includes the 8t for training AI and the 8i for running it. The company says nearly 100 customers now use its TPUs, including frontier AI labs, trading firms, retailers and universities. Alongside the chips Google also open-sourced TPU Sync, a tool it says cuts the wait before an AI starts replying by more than 90%.
WHY IT MATTERS
Nvidia’s real power over your AI bill comes from lock-in, because every AI company built on its chips faces a costly rewrite to leave. Google is attacking that lock head on, and the companies behind every app you use gain bargaining power the moment switching chips gets easy. So when your chatbot quietly raises its free limits or adds memory without a price hike, this chip fight is a big reason why.
WATCH TIP
What if AI could create entire virtual worlds you can actually explore? Google's new Genie 4 takes AI-generated video to the next level, turning simple prompts into interactive 3D environments that respond to your actions in real time. This short demo shows why AI-generated worlds could change the future of gaming, robotics, and beyond.
03 / SMALL AI
Source: Meta
Evolving AI: Muse Ember is Meta’s smallest AI, built for the repetitive tasks agents run all day.
KEY POINTS
Anthropic’s Claude Haiku 5.5 matches Ember’s top score while costing about three times more per task.
Haiku still beats Ember on one coding test, 34.9 to 27.2, and loses most of the others.
Meta’s own Muse Spark 1.3 scores only one point higher on intelligence, yet each task on it runs 17 times the price.
DETAILS
Meta launched Ember, and it scored 37.8 on an office workflow test to beat OpenAI’s GPT-6 Luna at 20.2 and Claude Haiku 5.5 at 13.9. It is the first model Meta trained with Muse Wonder Full and can take in a million tokens of text or video at once. Pricing sits at 10 cents per million input tokens and 50 cents per million output tokens for most prompts. Developers can use Ember through Meta’s Model API and Muse Code plus Google Cloud and Oracle Cloud.
WHY IT MATTERS
Anthropic and OpenAI now face real pressure to cut prices on their own small models as developers compare costs. When AI gets cheaper for the companies building apps, the tools you pay for each month tend to add more features without raising what they charge you.
04 / HUMAN OR MACHINE?
One is real. Trust your eyes?
Two images. One photograph. Click the one you think is real.
A
B
05 / QUICK HITS
🇦🇺 Australia announced plans for a new AI safety regulator and mandatory incident reporting, following an OpenAI agent’s unauthorized access to a government health statistics portal.
🦾 NVIDIA introduced GR00T 2 and Isaac Lab 3.0, open tools to train robots in simulation for precise tasks such as handling cables and assembling components.
🚨 OpenAI revealed Control Room, a shutdown system for internal AI research, which it says can block further model output within 35 seconds of a stop command.
🗺️ Google added Maps Platform directly to Google AI Studio, letting developers prototype map-based apps without manually setting up cloud credentials.
06 / TRENDING AI TOOLS
🗣️ Wispr Flow - Voice-to-text AI that turns speech into clear, polished writing in every app*.
💬 DocsAlot - Answers customer questions in Slack and Discord from your docs.
🔐 Pinku - Runs private AI chats fully offline on iPhone and Mac.
🧠 Turnstone - Runs personal AI agents using the AI subscriptions you already have.
*PARTNER LINKS
BEFORE YOU GO
Worth your four minutes?
Help us make the next edition better.
See you tomorrow,
The Evolving AI team
